mine-detection

Mine-detection is a custom ROS 2 pipeline designed for autonomous landmine detection using an Unmanned Aerial Vehicle (UAV). It utilizes a Software Defined Radio - PlutoSDR, acting as a Vector Network Analyzer (VNA) to transmit and receive radio frequencies. The system processes the received complex IQ data in real-time to detect anomalies in the soil dielectric properties and fuses these detections with the drone's odometry to accurately log the physical locations of detected mines

Table of Contents

• Quick Start
• Modules and Architecture
• Parameters

1. Quick Start

This project has been developed and tested on Ubuntu 22.04 (ROS 2 Humble). However, it should be compatible with other versions of Ubuntu and ROS 2 as well.

Firstly, to install the required (system+python) dependencies:

sudo apt-get update && sudo apt-get install -y \
  libiio-dev \
  libgpiodcxx-dev \
  libomp-dev \
  libeigen3-dev

pip3 install numpy pandas scipy matplotlib PyQt6

Then simply clone and compile the package within your ROS 2 workspace:

cd ~/ros2_ws/src
git clone https://github.com/AerialRobotics-IITK/mine-detection.git
cd ~/ros2_ws
colcon build --packages-select mine_detector

After compilation, source the workspace and launch the nodes:

source install/setup.bash

# Run the VNA detector node
ros2 run mine_detector mine_detector

# Run the localization and recorder node
ros2 run mine_detector mine_recorder

2. Modules and Architecture

The core algorithms for signal processing, spatial localization, and analysis are implemented in the src/mine_detector/src directory:

• mine_detector (port.cpp): This is a dual-threaded C++ ROS 2 node that interfaces with the PlutoSDR hardware via libiio. It handles real-time signal acquisition and processing. The node captures raw IQ samples, applies a precomputed lock-in amplifier algorithm utilizing OpenMP, calculates the S21 parameters, and performs signal smoothing using Gaussian and Savitzky-Golay filters. By dynamically calculating a baseline and thresholding the differential, it robustly flags the presence of targets in the soil and publishes the detection timestamps.

• mine_recorder (mine_recorder.cpp): This node synchronizes the detections from the VNA module with the drone's position. It tracks the drone's trajectory by subscribing to the odometry/pose topics. When a mine detection timestamp is received, it verifies the drone's stability (to prevent false positives during sudden accelerations) by interpolating historical velocities within a specified window. Upon confirmation, it logs the mine's 3D coordinates, publishes them to /detected_mine_pos, which is then subsequently relayed to the navigation drone swarm.

• solverc.py: A comprehensive Python-based Qt GUI application designed for post-flight analysis. It ingests the CSV log files produced by the mine_detector node, providing rich interactive visualizations of the Raw S21 data, Savitzky-Golay filtering, baseline tracking, and differential signals. It automatically highlights the exact intervals where the drone detected a mine and annotates the pinpointed local minima corresponding to the mine's center.

3. Detection Principle (S-Parameters)

The detection pipeline relies on the analysis of Scattering parameters (S-parameters), which quantify how RF energy propagates through a multi-port network.

• S11 (Return Loss): Represents the ratio of power reflected back to the transmitting antenna due to impedance mismatches.
• S21 (Forward Transmission / Insertion Loss): Represents the power successfully transmitted from the Tx antenna through the medium and received by the Rx antenna.

Target Isolation: In this system, the soil acts as the primary propagation medium. Under normal conditions, the S21 signal establishes a relatively stable baseline. The introduction of a buried landmine (metallic or non-metallic) creates a sudden dielectric contrast within the soil volume. This dielectric anomaly causes localized scattering and absorption of the RF waves, manifesting as a distinct, measurable deviation in the S21 transmission magnitude. By continuously tracking S21 and isolating these deviations from the baseline, the system can reliably distinguish anomalous targets from homogeneous soil.

4. Experimental Analysis and Methodology

The frequency selection and segregation logic were empirically derived by mounting the antenna array on a mobile testing rig (a suitcase setup) and traversing varied terrain profiles with known buried targets.

Data collection utilizing the mobile test rig.

Frequency Band Selection

Initial evaluations conducted at lower frequency bands (1 GHz to 1.4 GHz) yielded inconsistent detection profiles. The differential (|Diff| dB) between the background soil and the target was minimal (approximately 2 dB), making it difficult to separate true positives from environmental noise and natural soil variations.

Inconsistent S21 differentials observed at 1 GHz, 1.2 GHz, and 1.4 GHz.

Subsequent spectral analysis indicated that stepping up to the 2 GHz to 2.2 GHz band provided optimal contrast for the specific soil-target dielectric mismatch. At these frequencies, the presence of a mine induced a sharp, unambiguous 6 dB to 8 dB differential in the S21 signal magnitude.

Pronounced 6 dB to 8 dB differentials isolated within the 2 GHz band.

Signal Segregation Logic

Based on the 2 GHz band data, the real-time detection strategy was formalized as follows:

1. Signal Conditioning: Raw S21 samples are smoothed using cascaded Gaussian and Savitzky-Golay filters to attenuate high-frequency noise.
2. Dynamic Baselining: A rolling median estimator tracks the slow-varying soil baseline.
3. Threshold Detection: The absolute difference |Diff| between the conditioned S21 signal and the baseline is computed. If |Diff| exceeds the configured detect_thresh (e.g., 8.0 dB), a target is flagged. The local minima within this thresholded region precisely corresponds to the physical center of the target.